TW540008B - Color image processing method and apparatus thereof - Google Patents

Abstract

A color image processing method for retrieving a color feature descriptor for describing color features of an image is provided. The color image processing method includes the steps of (a) obtaining color vectors of an input image, (b) classifying the color vectors to obtain dominant colors of the input image and the ratios thereof, and (c) representing the dominant colors and the ratios thereof as a color feature descriptor of the input image. The color image processing method is applied to an object-based image processing method, thereby allowing fast search and retrieval of multi-media contents.

Description

540008 V. Description of the invention (1) Related applications refer to the right of this application for the benefit of US Provisional Application No. 60/1 1 8, 742 filed on February 5, 1999 The disclosure is incorporated herein by reference. BACKGROUND OF THE INVENTION 1. Scope of the Invention The present invention relates to a color image processing method and device, and more particularly to a color image processing method for capturing color feature descriptors used in color image indexing and searching. 2 · Related technical descriptions Among the various visual features used to describe multimedia content, color is the most advantageous feature. According to the conventional color image processing method, a color shape histogram is used to represent color information of an image. However, the conventional color image processing method using a color histogram including a storage area of 102 is disadvantageous in that it describes the image processing steps of the image, and its computational complexity requires a lot of processing time. ° Summary of the invention, to solve the above problems, the purpose of the present invention is to provide a color image processing method that can reduce the heterogeneity and processing time by 50%. Another object of the present invention is to provide a computer-readable, brain-executable program for performing color image processing, which has a further object. Another object of the present invention is to provide a color performing color image processing method. 1 image processing § also prepared ... color image processing method to realize the features of the present invention,

〇 Following the rhyme

8902l03.ptd 540008 V. Description of the invention (2) (a) Obtain the color of an input image ..., ... ^ to obtain the main color and Rate, and color = quantity classification The main colors and their ratios, as shown in (c) These color vectors are preferably quantities; feature descriptors. Digits. Qianhua Xiangxia, and its ratio is preferably a percentage. The color image processing method may further include step (e)...... Lizi images. This is the same, and the results are not combined as a whole—the same as before step (b), step 1 can further include a step to execute a pre-set and then soften the input image. Alternatively, the method may further include a step of performing a predetermined filtering process in step (?) To remove noise from the input image. At the same time, before step (b), the method may further include two steps: analysis The day pixels (pixels) in the filtered image are the probability of noise pixels, and appropriate weighting is given to them; and the corresponding weighted pixels are subjected to general Lloyd's algorithm (general L1〇yd alg) 〇rithm) to perform color quantization.-According to another aspect of the present invention, a color image processing method is provided to capture color feature descriptors describing the color characteristics of an image. This method includes the following steps: (a ) Divide an input image into a plurality of regions, (b) obtain the color vector of each divided region, (c) classify the color vector to obtain the main colors and ratios of the input image, and (d) indicate these main Colors and their ratios are used as descriptors of the color characteristics of the input image. This service also provides an electronic thin media, which has a computer executable

8902103.ptd Page 5 540008 5. Description of the invention (3) is used to execute a color image processing method * to capture the description. 2) Jiang: The color feature descriptor of the feature. This method includes the following steps: θ Ximing I turns the image into multiple regions, (b) obtains each divided region 1 4 t A to 1, (C) classifies the color vector to obtain the master and slave helmets of the input image 2 and its ratio 'and (0) represent these main colors and their ratios as color feature descriptors of the input image. According to another aspect of the present invention, a color image processing device is provided for capturing color feature descriptors describing color characteristics of an image. The device includes a color vector acquisition unit and a color feature descriptor generation unit. The color vector extraction unit receives daytime value data of an input image, and extracts a shape vector to a predetermined color coordinate system. The color feature descriptor is generated after all color vectors have been received. The color vector represents the major color percentiles, and generates and outputs color feature descriptor data containing information about the major colors and their percentiles. The present invention also provides a color image processing device for capturing color feature descriptors that describe image colors and features. The device includes a segmentation unit, a color f-vector acquisition unit, and a color feature descriptor generation unit. The segmentation unit divides an input image into k regions (where k is an arbitrary positive integer t) ′ and sequentially outputs day value data corresponding to the kth region. The color receiving unit 5 receives pixel value data of an input image, and extracts a color vector to a predetermined color coordinate system. The color feature descriptor generation unit obtains the main color percentiles represented by the color vectors after all color vectors have been received, and generates and outputs color feature descriptor data including the main color and its percentile information .

8902103.ptd Page 6 540008 V. Description of the invention (4) The preferred embodiments illustrate the above objects and advantages of the present invention, which can be more clearly understood by referring to the detailed embodiments of the drawings. The drawings include: Embodiments of the present invention will be described in detail with reference to the drawings. Please refer to FIG. 1, which shows a color method according to the present invention. First, a color image A is input (step 100). This color image is processed into a plurality of areas Fl, f2, F3, and ^ (step 102). This = image segmentation is performed based on, for example, edge f 1 ow. After the deletion, the knife cut can be called the quantized color direction of 卩 2 ~ and 卩 4; (The steps of obtaining the quantized color vector in each step and the center of the heart, preferably include first, performing-a predetermined filtering step to soften and Removal of the ζ step: the first pre-processing step. Secondly, the probability of m in the filtered image as noise is the daylight element, which is from the class with the adjacent daylight element, here the book color; Separate the pixels, and 1 color = in the selected pixels, and the prime value sighs as the maximum color distance, expressed by τ (η). Then T (0 weights the color vector of each day element. ExP (-T (n)) is the meaning of ▽. The stomach is next 'assuming that the average number of T (n) values of all daylights is Ta' to be used in the number of initial clusters of quantization. = An arbitrary constant, such as 2. Then, a general Lloyd's algorithm is performed on the color shirt vectors corresponding to the weighted = ^, and these color vectors are first used as the cluster centroid (ci) represented by the heterogeneous (1):

8902103.ptd Page 7 540008 V. Description of the invention (5) ^ Σν (η) × (η) / Σν (η) ··· _ · (1) where 'x (η) is the η Diurnal value of a diurnal element. Then calculate the Di value represented by the formula (2): I = Σ V ⑻ 丨 X⑻- "丨 丨 2 ··· _ ·· (2) Then divide the cluster with the largest Di value. Repeat this procedure until N clusters are generated After the N clusters are generated, the general Lloyd algorithm is executed. When the general Lloyd algorithm is executed, the cluster centroid is calculated using equation (1) to perform updating. Secondly, 'agglomerative clustering' is performed to Condensed clusters with the same color vector. Condensed clusters are composed of D. Duda and ρ · E ·

Hart is disclosed in the book "n Classification and Scene Analysis" published by John Wiley and Sons, New York, 1973. This book will not describe this book in detail. Then 'classify color vectors , And obtain the color vector [c L, cu, cVJ represents the main color and its percentile Pi (step 106). Here, i represents any serial number of the main area, ranging from! To] ^; L, U & v represent the coordinates of the cie LUV color coordinate system. The percentile Pi is expressed by decimals. The sum of the percentiles Pi of the ^^^ area is 1, as represented by formula (3): | ΣΡ, = ΐ [... · __ (3) Secondly, the main color percentile Pi: pi represented by the color vector [cLi, cUi, cVi] is represented as a color feature descriptor of a relevant area: cattle 108). In other words , The color feature descriptor F can be expressed by the formula ⑷:

540008

Bu = {{[cLi'cUi'cVihPi}, Ui, ..., N} ... ⑷, where N is a predetermined positive integer. The color feature descriptor can be called a variable storage area color histogram (variable-bin c0100r history). Combining the daytime prime data in the k-th region (that is, Regio) with the color feature descriptor data (that is, Fk) in the region, the image A 'can be expressed by the formula (5): A 丨 Reg i oni ,, Reg i 〇n2, F2; ......; Regionk, Fk} ......... (5) /, 'k 疋 a pre- 疋 positive integer, which represents the number of cut areas of image a (Step 110). θ The color feature descriptors extracted by the original and bright color image processing method. Second = each area is represented by a small number of numerical compression. Compressed representation of color specific description = compact (compact can significantly reduce the calculation complexity. This allows fast searching and capture of multimedia-based inner valleys. The color image processing method of the present invention can be applied to WMPEG-7 Object-referenced image processing method. The color image processing method of the present invention can use a computer program as a program to set the coding and coding segments (c odesegments) constituting the electrical program, which can be easily derived by related technologies ^ computer programmers. Programs are stored in electronic media, and can be read and executed by a computer, so as to specifically realize the shirt image processing method. The aforementioned media include magnetic recording media, optical recording media, carrier media, etc.-In addition to this The color image processing method can be implemented on a color image processing device. FIG. 3 is a block diagram of a color image processing device according to the present invention. Please refer to FIG. 3 'This color image processing device includes a segmentation unit 300

8902103.ptd Page 9 540008

,-A color vector acquisition unit 302 304 and a combination unit 306. A color feature descriptor generating unit When the color image processing device is in operation, the segmenting unit 300 divides the input image A into k regions' and sequentially executes the sequence ^ fReglnk. The color vector extraction unit 302 receives the query in the k-th area and manipulates the color vector [cLi, cUi, Xiu when the 4-way 篁 [cLpcUi'cVi] has been received. The main color percentiles P, 'represented by the color vector [cLi, cUi, cVi] are generated and output as the number of color feature descriptors. The color feature descriptor data Fk includes information about the main colors and their percentiles Pi represented by the color vectors cUi, cVJ. In order to obtain the percentile Pi of each color, it is best to perform ^ color == in each divided area. Therefore, the color image processing apparatus preferably further includes 7L (not shown). The color image processing apparatus preferably further includes a filtering unit (not shown) for performing a predetermined pass sequence to soften the input image and remove noise. The quantization unit analyzes the filtered and filtered images with a probability of several pixels and noise, and gives them appropriate weighting, and uses a general Lloyd algorithm to quantize the color vector corresponding to the weighted day pixels. Into. 306 combines the day value data in the k-th area (ie, Regi0r ^) with the color feature descriptor data (ie, Fk) in this area to output the processed image. The color image processing apparatus of the present invention can be applied to an object-based image processing method such as MPeg-7. In addition, in the color image processing apparatus of the present invention, a method for representing a color image using the main color of the image, in addition to color

540008

8902103.ptd Page 11 540008 Brief description of the diagram Brief description of the diagram FIG. 1 is a flowchart showing the color image processing method of the present invention; FIG. 2 shows the image segmentation performed in step 106 of FIG. 1; 3 is a block diagram of the color image processing apparatus of the present invention; and FIGS. 4A and 4B show the results obtained by performing a partition search on an image indexed by a computer program according to the color image processing method of the present invention. Graphic component description 3 0 0 segmentation unit 302 color vector extraction unit 3 04 color feature descriptor generation unit 3 0 6 combination unit

8902103.ptd Page 12

Claims (1)

540008 ~ I ♦ 正) ί · ί〇. 2 4 7 8 A color image processing method for capturing color feature descriptors describing the color characteristics of an image. This method includes the following steps: (a) obtaining an input map The color vector of the image; (b) classifying the color vector to obtain the main colors of the input image and their ratios; and (c) representing the main colors and their ratios as the color feature descriptors of the input image. For example, the color image processing method of the scope of application for patent No. 1 wherein the color vectors are quantized color vectors. For example, the color image processing method of the first patent application range, wherein the net ratio is a percentile. For example, the color image processing method in the first patent application scope further includes step (d): combining the color vectors and the color feature descriptor ′ and expressing the combined result as an overall image. For example, the color image processing method according to item 4 of the application, wherein the color vectors are quantized color vectors. For example, the color image processing method in item 4 of Shengu's patent scope, wherein the specific ratio is a percentile. For example, the color image processing method according to item 2 of the patent application scope further includes step (e): combining the quantized color vectors and the color feature descriptor, and expressing the combined result as a whole image. The color image processing method of the first item in the m range, in step (b), further includes a step of performing a predetermined filtering process to soften the input image 2002 · 10 · 31. 014 54〇〇〇8 said amended ^^ Case No. 89103QQS___ Year ^, applied for patent scope 9 · If the color image processing method of the first scope of the patent application, before step (b), It further includes a step of performing a predetermined filtering process to remove noise in the input image. 10. The color image processing method according to item 1 of the scope of patent application, before step (b), further includes a step of performing a predetermined filtering process to soften the input image and remove noise from the input image. 11 The color image processing method according to item 1 of the scope of patent application, before step C b), further includes the following steps: In the filtered image, the probability that the pixels are noise pixels is several. It is appropriately weighted; and quantization is performed on the amount of color using a weighted diurnal corresponding to a general Lloyd algorithm. Μ 2-a color image processing method for capturing color feature descriptors describing the color multiplication of an image, the method includes the following steps: (a) segment an input image into a plurality of regions; (( Color and b) obtain the color vector of each segmented area; c) classify the color vector to obtain the Gan Gangu of the input image > · ____ work sound color (d) represents the main colors and their ratios as the color characteristics Descriptor. The color vector is the quantized color vector as described in the color image processing method of item 12 of the patent application. Among them, 4 is the color image processing of item 12 of the scope of patent application. These ratios are percentiles. ' 'among them 540008 I Case No. 8910 ^ 998 _year_month said you 1 六 、 Scope of patent application 〆 ，， ·· V .； ': 丨 二 15. The color image processing method according to item 13 of the scope of patent application, including step (e): combining the quantized color of the segmented regions with the color feature descriptor 'and indicating the combination result as an error. For example, before step (b) of the color image processing method of the scope of application for a patent, step (b) further includes a step of performing a predetermined filtering shoulder to soften the input image. 17. The color image processing method according to item 12 of the patent application scope, before step (b), further includes a step of performing a predetermined filtering shoulder to remove noise in the input image. 1 8 If you apply for (b) to make the loser 19 If you apply for (b) to make each 20 like to apply for (b) to remove 2 1 if you want to apply for (b) so that each 2 2 like for (b) ) Before the patent scope, before entering the image soft patent scope, before dividing the area patent scope, before each division patent scope, and then before dividing the regional patent scope, it includes a step of performing a predetermined filtering and removing the input map. Noise in the image. The color image processing method of item 12 includes performing a predetermined pass softening in one step. The color image processing method of item 12 includes a step of performing noise in a predetermined domain. The color image processing method of item 12 includes a step of performing a predetermined filtering softening and removing ^ in each divided area. The color image processing method of item 12 includes the following steps: 〆 Further vector: image in step. , In step, step in step, step in step, step in step, step in step, 0 in step 540008 __ Cliff No. 8910 Qi QS__ Color VI. Scope of patent application Revision; negative positive 1 green -k- h X] ife El f-Fr r donated to the post-transition image, the probability of pretending to be a few, and giving it a proper prime is a noise pixel. The general Lloyd's algorithm is used to perform a quantum on a long amount. Into. Μ-weighted pixel-to-color direction 2 3 — computer-readable media, with a programmable code that can be inserted with a strong ~ Am electric & order '. ^. Shirt color image processing method for capturing Describe a color feature descriptor of an image color = sign. This method includes the following steps: (a) dividing an input image into a plurality of regions; (b) obtaining the color vector of each divided region; (c) dividing the color vector Classify to obtain the main colors and their ratios of the input image; and (d) represent the main colors and their ratios as color feature descriptors for the input image. 2 If the computer-readable medium of item 23 of the scope of patent application, further including step (e): combine the quantized color vector of the segmented regions and the color feature descriptor, and indicate the combination and the result as a whole image . 2 5 A color image processing device for tree color feature descriptors describing the color characteristics of an image. The device includes: a color vector capture unit for receiving day value data of an input image, and Extracting color vectors to a predetermined color coordinate system; and a color feature descriptor generating unit, when all color vectors have been received, obtaining the percentiles of the main colors represented by the color vectors, and generating and outputting Color and its percentile information 540008 Amendment 6 Case No. 89103998 _ Year House—Japan Patent Application Scope > Color Feature Descriptor. 2 6 = The color image processing device in item 25 of the Chinese Patent Application ^ includes a combination of modules to combine pixel value data and color feature data to output a processed image. Tian Shu 27—A color image processing device for capturing color feature descriptors describing an image feature. The device includes ... /, a two-segmenting unit for segmenting an input image into k regions. ,, Where k is an arbitrary positive integer, and then the pixel value data in order; the marginal domain symplectic value: the fetching unit 'is used to receive a pixel of an input image = data, and the color vector is given A predetermined color coordinate system has been received Π: Descriptor generating unit 'When all color vectors have been received, it receives the main number represented by the color vector, and generates and outputs color features including the main color and the hundreds of positions Descriptor color and its percentile information 2 8 Such as the scope of patent application No. 2 7 ^ a Including the image processing equipment of 7 items, and further subdivided. As early as 70, it is a color image processing device used to execute the color amount of each of the divided areas 2 9: 2 Item 28 of the patent scope. It is advisable that the profiling unit analyzes the filtered probability and gives it appropriate Adding; Medium = the color of the noise pixel to H ^ weight and then corresponding weighted pixel pairs. Light and Lloyd's algorithm to perform color quantization 890210301.ptc page 17 2002, 10.31.018 540008 _ case number 89103998 _ year month and month revision _ 6, the scope of patent application 30, such as the scope of patent application number 27 The color image processing device further includes a combination unit for all k divided regions, combining pixel value data corresponding to the k-th region and a color feature descriptor of the corresponding region to output a processed image. 31 The color image processing device according to item 30 of the patent application scope further includes a quantization unit for performing color quantization in each divided region. 32 2 The color image processing device according to item 31 of the scope of patent application, wherein the quantization unit analyzes the probability that the pixels in the filtered image are noise diurnal elements, and gives them appropriate weighting, and then weights accordingly. The pixel performs a general Lloyd's algorithm on the color vector to perform color quantization. The color image processing device, such as the 27th item of the patent application scope, further includes a filtering unit for performing a predetermined filtering. Processing to soften the input image. The color image processing device according to item 27 of the patent application scope further includes a transition unit for performing a predetermined filtering process to remove noise in the input image. The color image processing device according to item 27 of the patent application scope further includes a filtering unit for performing a predetermined filtering process to soften the input image and remove noise in the input image. The color image processing device according to item 33 of the patent application scope further includes a quantization unit for performing color quantization in each divided region. 890210301.ptc Page 18 2002.10.31.019 540008 Sixth 3 ^ 8910 ^ QR patent application scope is amended 3 3 4 4 4 7 If the color scope of the patent application item 34 is included-the quantization unit is used to implement the map The image processing equipment is further subdivided.量 The amount of color in each segmented area = The color in item 35 of the patent scope includes a quantization unit for performing the v v processing Λ preparation and then sub-quantization.的 The amount of color 9 in each segmented area is as described in item 36 of the patent application scope. The quantization unit analyzes the filtered image = ^ all probability of the processing equipment and gives it an appropriate weight. 素: Noise: The prime color vector is subjected to general Lloyd's function to correspond to the weighted pixel pairs. XOR 'to perform color quantization of the 37th color image processing device, where the probability that the pixels in the post-image are the colors of the noise pixels' vector is performed like a Lloyd algorithm to perform color Color image processing equipment with quantized inn range of f38 terms, in which the probability is given appropriately: a general Lloyd for the color vector of the noisy day prime is then corresponding to a weighted pixel pair. "German / German method" to perform color quantization 2 such as the scope of patent application No. 3 9 # Mostly include-combination unit, oblique color image processing equipment, and then the G value of the kth area: all k division Area, combined corresponding ^ According to the color characteristics of the corresponding area 890210301.ptc 苐 Page 19 2002.10.31.020 540008 Case No. 89103998 Amendment $ iiiv .asl · * ',. \ .u ft Lai: 1 " ^ Γ, Γ ^ 7 ^ Six, the scope of patent application, To output a processed image. 4 3 The color image processing device according to the scope of patent application No. 40, further comprising a combination unit for all k segmented regions, combining the daytime prime value data corresponding to the kth region and the color feature description of the corresponding region ία 3 tilde to output a processed image. 44. The color image processing device according to item 41 of the scope of patent application, further comprising a combination unit for all k divided regions, combining the day-time value data corresponding to the kth region and the color feature descriptor of the corresponding region, To output a processed image. 4 5 —A method for representing a color image, the method includes the steps of: (a) obtaining the main color of the input image and its percentile; and (b) using the obtained main color and its percentile Represents the color image. 890210301.ptc Page 20 2002.10.31.021